Children's food preferences are learned, and familiarity is the strongest known influence on them (Birch, 1999). Repeated tasting of an initially disliked vegetable increases liking and consumption in experimental studies (Wardle et al., 2003). A garden offers exposure of a particular kind: children see the plant grow, handle it, and taste it in the company of classmates and teachers. Watching peers and adults eat a food, and gaining confidence in preparing it, plausibly make trying it more likely. Early classroom studies in the United States found that garden-enhanced nutrition lessons improved knowledge and preference for some vegetables (Morris and Zidenberg-Cherr, 2002). None of this guarantees a change in intake at the school meal, which also depends on what is served.

A school garden joined to a feeding programme is a complex intervention in the sense used by the Medical Research Council: it has several interacting components, relies on the behaviour of those who deliver it, and is sensitive to local conditions (Craig et al., 2008). The Council's guidance on process evaluation asks three things of such interventions: what was implemented, through what mechanisms it produced change, and how context affected both (Moore et al., 2015). Case study method is suited to exactly these questions. Yin (2018) defines a case study as an empirical inquiry into a contemporary phenomenon within its real setting, relying on multiple sources of evidence that are brought to converge, and guided by propositions stated in advance.

The evaluation begins from a programme logic that runs from garden to plate. Figure 1 sets out the five links and the evidence proposed for each. The central proposition is that vegetable intake at the school meal rises only when every link holds. Three rival explanations are written down at the start, as Yin recommends, so that evidence can be sought for and against them: that any change in intake reflects seasonal market supply and not the garden; that it reflects a change in the feeding programme's menu or budget; and that children's apparent liking is a response to being observed.

The case is one rural primary school that operates a cooked midday meal and a garden intended to supply it. Within the school, three embedded units are studied: the garden, the kitchen, and the pupils of two or three selected classes. A single case is justified when it is typical of a wider class of settings or when it offers unusual access (Yin, 2018). Selection should therefore be documented against stated criteria, such as an established feeding programme, a garden already in its second season, and a head teacher willing to open records. The school's size, water source, land and staffing should be described fully enough for readers to judge how far it resembles schools they know.

Table 1 lists the sources of evidence, what each records and the link in the logic it serves. A garden and kitchen log, kept by a teacher and the cook, records what is harvested, in what quantity, and where it goes. Periodic site visits by the evaluator check the log against what is seen. Structured observation of garden lessons records what pupils do. Group discussions with pupils and interviews with the cook, teachers and parents explore liking, habit and obstacles, and are analysed thematically (Braun and Clarke, 2006).

Intake is the hardest link to measure. Young children cannot report their own diet accurately, and recall methods that work for adults are unreliable below about eight years of age without a parent's help (Livingstone et al., 2004). The design therefore measures consumption at the school meal by direct observation of plate waste: trained observers record the portion of vegetables served and estimate the fraction remaining on each child's plate, a method whose agreement with weighing has long been documented (Comstock et al., 1981). Weighing a sample of plates on each observation day provides a check on the visual estimates. Observers need joint training and periodic calibration against one another for the estimates to be comparable.

Analysis follows a convergent mixed methods logic in which quantitative and qualitative strands are collected in parallel and merged at interpretation (Creswell and Plano Clark, 2018). For each link in the programme logic the evaluator sets out what would be seen if the link held and what would be seen if it did not, then compares the evidence with both patterns. The log and plate waste records are arranged as a time series across the school year, so that periods when garden produce reached the kitchen can be set against periods when it did not. If vegetables served and eaten rise and fall with garden supply, and market prices or menu changes do not account for the pattern, the central proposition gains support. If intake is flat whatever the garden yields, the break in the chain can be located.

The report should allow a reader to trace each conclusion back to the records on which it rests. Yin calls this maintaining a chain of evidence, supported by a case study database kept separate from the narrative. A written protocol, prepared before fieldwork and covering the questions, instruments, observation schedule and consent procedures, makes the study repeatable in another school. Draft findings should be returned to the head teacher, cook and garden teacher for comment on factual accuracy before they are finalised.